a a high frequency sampling monitoring system for

A

A high frequency sampling monitoring system for environmental andstructural applications. Part A: The hardware architecture

CESARE ALIPPI, ROMOLO CAMPLANI, CRISTIAN GALPERTI, ANTONIO MARULLO,MANUEL ROVERI, Dipartimento di Elettronica e Informazione, Politecnico di Milano, Italy

High frequency sampling is not a prerogative of high energy physics or machinery diagnostic monitoringonly: critical environmental and structural health monitoring applications also have such a challenging con-straint. Moreover, such unique design constraints are often coupled with the requirement of high synchro-nism among the distributed acquisition units, minimal energy consumption and large communication band-width. Such severe constraints led scholars to suggest wired centralised monitoring solutions, which haveonly recently been complemented with wireless technologies. This paper suggests a hybrid wireless-wiredmonitoring system combining the advantages of wireless and wired technologies within a distributed high-frequency-sampling framework. The suggested architecture satisfies the mentioned constraints, thanks toan ad-hoc design of the hardware, the availability of efficient energy management policies and up-to-dateharvesting mechanisms. At the same time, the architecture supports adaptation capabilities by relying onthe remote reprogrammability of key application parameters. The proposed architecture has been success-fully deployed in the Swiss-Italian Alps to monitor the collapse of rock faces in three geographical areas.

Additional Key Words and Phrases: Monitoring system, wireless sensor network, high-frequency sampling,rock collapse forecasting.

ACM Reference Format:

Alippi, C., Camplani, R., Galperti, C., Marullo, A., Roveri, M., 2012. A high frequency sampling monitoringsystem for environmental and structural applications. Part A: The hardware architecture ACM Trans. Sen-

sor Netw. V, N, Article A (January YYYY), 31 pages.DOI = 10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000

1. INTRODUCTION

Challenging environmental and structural monitoring applications, e.g., those relatedto rock collapse forecasting and structural health assessment, require high samplingrates to detect asynchronous events or events characterized by high frequency compo-nents. Quite often, the requested sampling rate is a few kHz, a rate introducing tech-nological challenges whenever low energy consumption, remoteness of the deployment,distributed data synchronization, power availability and large bandwidth constraintsare also requested.

The first monitoring systems for medium to high sampling rates were based on cen-tralised solutions [Boller et al. 2009] [Farrar and Worden 2007] with sensors directlystemming in a star topology from the data acquisition board. Such a configuration al-

This article extends the work published in the Proceedings of the IEEE 7th International Conference onMobile Adhoc and Sensor Systems (MASS), 2010.The research activity is partly funded by the INTERREG EU project M.I.A.R.I.A. (An adaptive hydrogeo-logical monitoring supporting the alpine integrated risk plan) Italy-Switzerland action 2007-2013, and theKIOS Cyprus-funded project.Authors addresses: C. Alippi, R. Camplani, C. Galperti, A. Marullo, M. Roveri, Dipartimento di Elettronicae Informazione, Politecnico di Milano, Italy. Email: alippi, camplani, galperti, marullo, [email protected] to make digital or hard copies of part or all of this work for personal or classroom use is grantedwithout fee provided that copies are not made or distributed for profit or commercial advantage and thatcopies show this notice on the first page or initial screen of a display along with the full citation. Copyrightsfor components of this work owned by others than ACM must be honored. Abstracting with credit is per-mitted. To copy otherwise, to republish, to post on servers, to redistribute to lists, or to use any componentof this work in other works requires prior specific permission and/or a fee. Permissions may be requestedfrom Publications Dept., ACM, Inc., 2 Penn Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212)869-0481, or [email protected].© YYYY ACM 1550-4859/YYYY/01-ARTA $10.00DOI 10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000

ACM Transactions on Sensor Networks, Vol. V, No. N, Article A, Publication date: January YYYY.

A:2

Fig. 1. The high-level architecture of the proposed hybrid monitoring system

lows the monitoring system to have a high sampling rate and data synchronism amongunits. Unfortunately, the peer-to-peer wired solution limits the number of probes andthe deployment area covered. The length of the cable increases signal attenuation, lim-its communication bandwidth and requires a long length of cables. More flexibility canbe obtained by considering wireless solutions at the cost of higher energy consumption.

Herein, we propose a hybrid wireless-wired solution where advantages of wired andwireless technologies are combined to suitably address applications requiring highsampling rates. The systems approach depicted in Figure 1 presents a WSN structurewhere each node is a hybrid remote monitoring system. Sensing units are connectedthrough a fieldbus and controlled by a unit exposing a wireless communication capa-bility (base station). The base station provides energy to connected units, a local highcommunication bandwidth, strict time synchronisation and a remote wireless commu-nication channel to deliver data within a mesh network configuration.

The main novel contributions of the proposed system can be summarised as:

— presentation of systems design where WSN units are the coordinators of a fieldbus-based wired network of sensors and actuators;

— a unique design for a distributed embedded system where units consume low-power(about 300mW at the 2kHz sampling rate);

— design of hardware supporting advanced hierarchical energy management policiesand up-to-date energy harvesting mechanisms for low-power photovoltaic cells oper-ating in a non-stationary environment;

— introduction of fault robustness mechanisms at the unit level;— possibility to send commands to the remote system to change the operational mode

of the electronics.

This article extends a preliminary version of the work published in [Alippi et al.2010]; the technical material will be presented in two companion papers. The firstpaper, mainly devoted to hardware and architectural aspects, addresses all issues re-lated to the hardware architecture, the designed embedded system, the fieldbus andthe energy harvesting mechanism; the second manuscript targets at software aspectsincluding data communication protocols, unit management, remote web software inter-face, control room design and data interpretation. The companion paper can be found


A:3

in [Alippi et al. 2012]; this paper addresses the hardware aspects and is structuredas follows. Section 2 critically analyses and contrasts available monitoring systemsfor high-frequency sampling found in the literature. Section 3 introduces the applica-tion requirements and the architecture of the hierarchical wireless-hybrid monitoringsystem proposed. Section 4 presents the sensor and processing units; base station de-sign and CANbus aspects are detailed in Section 5. Section 6 focuses on the systemenergetic design. Finally, Section 7 describes the system applied to a real-world de-ployment, specifically designed to solve a rock collapse forecasting application.

2. STATE OF THE ART

There are several monitoring systems dealing with very high sampling rates, mostlyassociated with particle physics experiments or industrial monitoring; as an example,the interested reader can refer to [Dell’Acqua et al. 1993] where a read-out system forhigh energy physics is presented, or [Gungor and Hancke 2009; Ramamurthy et al.2007] where high frequency sampling is used to detect faults and design preventivemaintenance activities. Interestingly, a high-frequency distributed measurement sys-tem for industrial monitoring purposes is described in [Schmid et al. 2010]; herein, thegoal is to detect changes in the vibrational pattern of machineries to predict the needfor maintenance.

High frequency sampling for distributed monitoring systems has been primarily con-sidered in environmental monitoring applications, e.g., those addressing landslides,unstable cliffs, rock faces and volcanic eruptions. More specifically, [Werner-Allen et al.2006] proposed a wireless system for monitoring volcanic eruptions. Therein, 16 sen-sor nodes continuously sample seismic (single-axis and tri-axial seismometers) andacoustic data (through a unidirectional microphone) at 100 Hz. The system adoptsthe Flooding Time Synchronization Protocol [Maroti et al. 2004] to keep units syn-chronised. Unfortunately, the absence of energy harvesting mechanisms limited theoperational lifetime of the network to 19 days. [Glaser 2005] proposed a monitoringsystem constisting of down-hole seismographic arrays. Each array comprises a num-ber of independent sensing units (dispersed in the borehole) linked with a RS-485shared bus to a local gateway. Each unit (whose cost is approximately $6000) mountsa 16bit microprocessor fitted with a tri-axial MEMS accelerometer, tilt sensor, mag-netometer and pore pressure sensor. Data coming from the accelerometer, sampled at50 Hz, measure the seismic wave propagation, precious information for building upground-motion predictions. The Linux-based local gateway is remotely connected tothe Internet by means of a dedicated radio link. Authors claim to be able to guaranteesampling rates up to 250Hz. This constraint is imposed by the TinyOS operating sys-tem which, by adopting a run-to-completion scheduling mode, cannot guarantee anytime execution deadline for the sampling and processing phases. The power consump-tion aspect is critical here since each unit consumes about 400 mW (probably since theoperational setup is set to 50Hz) and the local gateway is always active (no duty-cycle)to provide the synchronisation clock tick to all bus-connected sensing units.

[Ingelrest et al. 2010] presents an interesting hardware/software architecture forenvironmental monitoring systems. Several deployments are presented and discussed(including the monitoring of a glacier and a bridge) but the sampling rate is low sincethe designed framework was designed for hydrological and micro-meteorological mon-itoring applications.

Different application scenarios for high-frequency data monitoring are presented in[Eisenman et al. 2009] and [Hu et al. 2009]. [Eisenman et al. 2009] suggests a mo-bile networked system for bicyclists aiming at measuring the cyclist performance andenvironmental data. The hardware platform is Moteiv Tmote Invent [Moteiv 2007],fitted with a two axis accelerometer (sampled at 115Hz) to measure the angle and


A:4

the lateral tilt of the bicycle. The prototyping units are powered by off-line recharge-able batteries without any energy harvesting mechanisms. [Hu et al. 2009] presents aWSN-based distributed system for monitoring the population of native frogs and inva-sive cane toad in northern Australia through an automatic recognition of the animalvocalization. Acoustic signals require high sampling rates (up to 10kHz) and complexprocessing. The proposed system is organized in two layers: low-power Mica2 motes(7.7MHz Atmega processor, 512Kb flash memory) [CrossBow 2009] acquire acousticsignals through microphones, locally process and transmit them to a Stargate board(400MHz Intel PXA processor, 64Mb RAM, 32Mb flash memory) for further process-ing and signal classification. Since the mote lifetime is approximately 4 days, authorsproposed the use of rechargeable batteries and solar panels but details about the en-visaged technological solutions have not been provided.

Structural Health Monitoring (SHM) applications [Farrar et al. 2011] [Boller et al.2009] [Grattan et al. 2009] require data to assess the structural properties of civiland aerospace infrastructures. Most of the time, accelerometers are embedded in thesensor platform to identify possible changes at the structural level (e.g., by evaluatingchanges in the natural resonance frequency [Farrar and Worden 2007]). The samplingrates of these systems generally range from 50Hz to 1kHz.

Traditional SHM systems are based on a centralised acquisition board which period-ically samples data from wire-connected probes. This solution guarantees synchronoussamples among different acquisition channels. However, the reduced length of theprobe cables and the limited number of signal acquisition ports limit the size of thedeployment area.

To overcome these limitations, [Caicedo et al. 2002; Engel et al. 2004] introduced aradio-based probe in which the signals acquired are directly sent to a central acqui-sition board by means of a radio module and no local processing is performed. [Lynchet al. 2004; Lynch et al. 2003] introduced the concept of smart sensors [Akyildiz et al.2002] in the SHM field. The sensing units proposed process locally acquired data (i.e.,through a FFT transform) and transmit remotely only the outcome of the computation(i.e., the relevant FFT coefficients); synchronisation aspects and energy harvestingmechanisms are not considered. [Lim and Shoureshi 2003; 2006] suggested a mon-itoring system for structural integrity of power lines based on the electromagnetic-acoustic transducer (EMAT) sensor [Johnson et al. 1994]. Sensing units observe thepropagation of electromagnetic pulses through the cable to detect potential structuraldamages (e.g., broken strands and/or corrosion). The signals acquired are locally fil-tered and classified by means of a neural network to identify the type of the damage;sensing units only act as data-loggers, since no transmission mechanism is provided.[Basharat et al. 2005] proposed a heterogeneous system consisting of a WSN and aset of wired cameras. The WSN nodes acquire signals from 2 MEMS accelerometers:the first one, which has a reduced measurement range, aims at detecting the naturaloscillations of the structure; the latter, characterised by a larger measurement range,is used to detect anomalous shocks. The events registered by both MEMSs are thenforwarded to the gateway for further processing (the gateway switches the cameras ononly if an event is detected). [Kim et al. 2007] suggested a wireless sensors networkto monitor the Golden Gate bridge. The peculiarity of this system is the in-line archi-tecture of the involved WSN nodes. A command sent from the control room activatesthe acquisition phase for all nodes, which store up to 20 minutes of measurementsto a flash memory. When the registration ends, nodes transmit data to the gateway.Curiously enough, this phase, which relies on a 46-hop protocol, requires 12 hours toreliably convey all the data acquired to the base station. A similar approach is pre-sented in [Rice and Spencer Jr 2009; 2008].


A:5

A wireless sensor network for structural health monitoring is presented in [Paeket al. 2005]. A set of Mica2 motes acquires tri-axial accelerometer measurements at50Hz and transmits them to a sink by exploiting a multi-hop architecture of the net-work. Details about energy harvesting or energy management mechanisms are notprovided. Interestingly, [Paek et al. 2010] re-implemented the system suggested in[Paek et al. 2005] through the proposed Tenet architecture. This novel multi-tieredarchitecture improves the scalability and flexibility of the system and allows for theremote updating of application parameters (e.g., the sampling frequency).

A critical comparison of the solutions presented above is summarised in Table Iand clearly shows that synchronisation, remote reconfigurability, and effective energyharvesting still remain open issues for high sampling rate monitoring systems. Unfor-tunately, these application requirements are not negligible in the design of effectiveand credible WSNs able to operate in harsh environments while guaranteeing a longlifetime of the network at those challenging sampling frequencies. In particular, syn-chronisation is essential for the analysis of high frequency events since even a min-imal time skew might introduce large errors e.g., in the detection and localisation ofa damage. The remote reconfigurability allows for fine tuning of the system after itsdeployment. This feature, which is essential for scenarios characterised by a sparseknowledge of the phenomenon under investigation, allows the designer to tailor thesolution parameters (e.g., filter taps and thresholds) to the application needs on-line.At the same time, it allows the system to track changes in the environment, e.g., dueto non-stationarity phenomena by adapting offsets, gains and other parameters rul-ing the system-environment interaction. Finally, energy harvesting mechanisms aremandatory for long life-time deployments since the battery would otherwise last onlya few days, given the high sampling rates. It is worth noting that effective and effi-cient mechanisms which are able to maximise the energy harvested from solar panelsworking in non-optimal conditions (e.g., presence of clouds and rain, dust and wateraffecting the solar cell) should be considered in real-world applications.

3. APPLICATION REQUIREMENTS AND SYSTEM DESIGN

The application requirements of strict synchronisation among units, remote reconfig-urability and energy sustainability pose a challenging design issue for the monitoringinfrastructure and the embedded electronics, which need to provide high performance(high communication throughput and sampling rate, strict real-time decision makingability and application code reconfigurability to track environmental changes) withinan energy hungry application in an environment of reduced energy availability. Thetrade-off between performance and constraints satisfaction was reached by consider-ing the hybrid wired-wireless monitoring system presented in Figure 1.

The monitoring system consists of one or more Remote Monitoring Sub-systems(RMSs) acquiring data from the environment and a control room whose goal is to re-trieve measurement data from the RMSs for storage and further processing. Sinceeach RMS monitors the environment independently of the others (units aimed at mon-itoring the same environmental niche are grouped in the RMS), we will describe asingle RMS. Note that the base station of an RMS provides the wireless mechanismfor building up a WSN, having each node as a terminal point of the wired-wirelessnetwork.

A generic RMS consists of the base station (BS) connecting a cluster of sensing andprocessing units (SPU) by means of a field bus (e.g., CAN bus). SPUs acquire andlocally process the gathered data (e.g., MEMS, geophones, temperature, strain gaugesensors) and provide a possible actuation action. The BS collects data coming from theSPUs and transmits them remotely to the Control Room (e.g., by means of a WiFi,UMTS or GPRS link).


A:6

Table I. Comparison tableHW platform Sensors Sampling

FrequencySynchroniz. Remote

Config.EnergyHarvesting

Local Pro-cessing

Seismographic ar-rays [Glaser 2005]

Custom unit tri-axial MEMS 50Hz Yes No Yes Yes

Volcanic eruptions[Werner-Allenet al. 2006]

Moteiv TmoteSky

single/triaxialseismometers,microphone

100Hz Yes No No Yes

BikeNet [Eisen-man et al. 2009]

Moteiv TmoteInvent

Two axis ac-celerometer,magnetic sen-sors, GPS

115Hz Yes No Yes/No Yes

Cane toad monitor-ing [Hu et al. 2009]

Mica2, Stargate Microphone 10kHz Yes No No Yes

SHM [Caicedoet al. 2002]

Analog board PiezoeletricMEMS

200Hz No No No No

SHM [Engel et al.2004]

Custom Tri-axialMEMS

100-650Hz No No No No

SHM [Lynch et al.2004]

Custom boardwith 8-bit AVR,32-bit PowerPc

Dual-axialMEMS

976Hz No No No Yes

SHM [Basharatet al. 2005]

MicaZ Tri-axialMEMS, tem-perature, videocameras

150Hz Yes No No Yes

SHM [Kim et al.2007]

MicaZ, customacquisitionoboard

Dual-axialMEMS

1kHz Yes No No Yes

SHM [Rice andSpencer Jr 2008]

Imote2, customacquisitionboard

Tri-axialMEMS, tem-perature

50-280Hz Yes No No Yes

SHM [Paek et al.2005].

Mica2 Vibration Card 50-160Hz Yes No No Yes

SHM [Paek et al.2010]

Mica2/MicaZ Vibration Card 50-160Hz Yes Yes No Yes

The remote control room stores and processes the data acquired (e.g., for filtering,pattern recognition, remote decision making). At the same time, the control room al-lows a human operator, or the system itself, to remotely change the parameters of theRMS through a simple web application by relying on decision making algorithms.

The BS provides both synchronisation among units and power supply to SPUs; atthe same time, a CAN bus solution guarantees a maximum skew among units below0.5ms (details can be found in the companion paper [Alippi et al. 2012]). This hybridarchitecture allows us to centralise the energy harvesting mechanisms at the BS level,which is endowed with solar panels and deployed in a safer, easily accessible, area(whereas SPUs generally are not in a safe area).

Design Rationale

The need to jointly address two opposite goals, such as high sampling acquisition ratesand long system lifetime, led us to consider a novel hybrid system architecture. Thishybrid wired-wireless solution aims at providing the advantages of the wired systemsin terms of energy efficiency and high synchronisation among the acquisition unitswhile maintaining the scalability and flexibility typical of wireless solutions.

In addition, such an architecture allowed us to design and develop SPUs withoutbatteries and solar panels since the energy harvesting mechanisms are centralised atthe BS. SPUs can thus be simpler, cheaper and smaller in size and weight. Deployedunits are required to be light and small in size since they are hung on the rock faceby means of holes and fixed with an ad-hoc resin. The absence of solar panels and,in particular, of the battery allowed us to significantly reduce both the size and theweight of the SPUs (see Section 4.4).


A:7

In such critical environmental conditions, the presence of solar panels on the SPUswould introduce two additional drawbacks. Firstly, vibrations induced by the wind onthe solar panels would propagate to the SPU case, thus affecting the acquisition phaseof geophones and accelerometers (e.g., inducing false events). Secondly, the orientationof the solar panels is very difficult, whenever applicable, thus reducing the energyharvesting efficiency.

In summary, the fully wireless solution, which is in principle feasible, would havesignificantly increased both the complexity and the cost of the overall system as wellas introduced possible bandwidth problems when bursts of events happen. Differently,a hybrid solution allows us to centralise both the energy harvester and the synchroni-sation issue, thus leading to a more credible and cheaper solution.

The choice of the CAN bus as fieldbus for the RMSs deserves a specific discussion.Although several well-known technological solutions are in principle available (e.g.,RS485, Industrial Ethernet, ProfiBus, ControlNet), we opted for the CAN bus for tworeasons. Firstly, the CAN bus is a widely accepted standard vendor free solution (e.g.,ProfiBus is a Siemens’s solution). For this reason, many microcontrollers and indus-trial PC104 boards are fitted with CAN bus interfaces. Secondly, the CAN bus providesboth the physical and the MAC layers of the communication stack, whereas other field-buses (e.g., the RS485) do not specify or suggest any communication protocol. Thisallowed us to simplify the design of the communication layer among the acquisitionunits.

Hereinafter in the paper we detail the main components of the RMS i.e., the SPU,the Base Station and the fieldbus; details related to the software are described in thecompanion paper [Alippi et al. 2012].

4. SENSOR AND PROCESSING UNITS

SPUs acquire measurements from sensors, carry out local processing to extract fea-tures from sensor data and transmit them to the BS. As presented in Figure 2,electronics-wise SPUs are organised in two boards and a set of sensors, i.e., straingauge, temperature, MEMS accelerometer, inclinometer and geophones. The SPUs areconnected to the BS by means of the CANbus which provides both the energy supplyand the transmission medium.

In more detail, each SPU has been designed as two vertically-stacked electronicboards, the first one carrying out processing and communication (see Section 4.1), thesecond one signal conditioning (see Section 4.2). The two boards are suitably tied to-gether with connectors and screws. The vertically-stacked architecture has been de-signed to reduce the size.

As detailed in Section 4.3, each SPU is fitted with a tri-axial MEMS accelerometerand a temperature sensor (to be used for thermal compensation and get indicationsabout the frost/defrost cycle). The MEMS accelerometer provides both accelerometerand inclinometer information. To provide a more detailed view of the physical phe-nomenon under monitoring, the SPU could be fitted with a geophone or a strain gauge.The geophone can provide information about the movements of the ground (in thiscase, the MEMS accelerometer only provides information about the inclination; ac-celerometer measurements are not acquired), while the strain gauge allows the defor-mation of a pre-existing fracture to be monitored.

An ad-hoc case, which is detailed in Section 4.4, has been designed and manufac-tured to reduce the overall unit size and dissipate thermal energy by convection, toease the deployment phase and avoid unwanted mechanical resonance of the casefalling in the 1kHz bandwidth. The cost of an SPU (boards, case and sensors) in this


A:8

Tranceiver

Power

CAN

DSPic33f

can vdd

12bit ADC

3

8

Lightning

protection

circuits

Lightning

protection

circuits

gp

io

DC/DC

3

dip

sw

itch

es

Inclinometer

signal amplification

Straine gauge


Temperature signal amplification

Microsesmic


vdd

MEMS

Temperature

Straine gauge

Geophones

3

3

Sig

na

l C

on

ditio

nin

g B

oa

rd

Pro

ce

ssin

g a

nd

co

mm

un

ica

tio

n b

oa

rd

Fig. 2. The architecture of the SPU.

low volume version is around 1000 euros. The cost would be significantly reduced witha large scale production.

4.1. The processing and communication board

The processing and communication board (see Figure 3a and 3b) basically consists offive main parts: a microcontroller, CAN transceiver, lighting protection circuits, dipswitches and DC/DC power regulator circuit.

Among the wide range of microcontrollers available on the market we opted for adigital signal processor (DSP) since it provides the hardware capability required tocarry out an efficient mathematical processing of a data stream (e.g., digital filteringand FFT).

In particular, we selected the ultra-low power DSP Microchip DSPic33F operatingat 40MHz and providing a simple 16-bit microcontroller architecture as well as an upto date two-bus DSP architecture with a 40 bit multiply and accumulate unit and twoseparate hardware address generators. The microcontroller fetches code for executionfrom the internal flash memory, uses an internal SRAM memory for data storage andprovides on-chip peripheral devices for external interfaces (such as CAN).

Based on our expertise, this ultra-low power DSP guarantees the best trade-off be-tween performance and power consumption. Off-the-shelf 16bit microcontrollers (e.g.,PIC18, MSP430, Cortex M0) do not provide DSP functionalities in hardware (whichmust be executed in software, thus increasing the power consumption of the SPUs).On the contrary, standard 32bit DSPs (e.g., Sharc by Analog Devices, the Texas In-strument C6000 or the Cortex-based STM32F4) provide higher computational perfor-mance at the expenses of an unacceptable power consumption in our applications (e.g.,the power consumption of these standard 32bit DPSs ranges between 500mW and2.5W in a fully active mode, while the envisaged 16bit DSP only consumes 300mW inthe same mode).

The basic tasks carried out by the DSPic33F are:

(1) Sampling sensor signals coming from the conditioning circuit;(2) Processing the digital data stream at 2k samples per second (for all connected

sensors requiring a high sampling rate);(3) Storing processed data to the internal SRAM memory.


A:9

Fig. 3. The boards composing the SPUs: a) Processing and communication board [upper side]; b) Processingand communication board [lower side]; c) Signal conditioning board [upper side]; c) Signal conditioning board[lower side].

The CAN transceiver is a circuit controlling the connection and data propagationalong the CAN bus. As such, it represents the physical layer of the CAN stack; theMAC and the network layers are handled by the microcontroller. We chose the TexasInstruments ISO1050 CAN transceiver since it provides the opto-isolation between theprocessing and communication board and the CAN bus (thus guaranteeing robustnessto overvoltage up to 4000V).

The lighting protection circuits aim at preventing the effects of lightings (in termsof overvoltage and overcurrent) that could fall in the neighbourhood of the system(the risk of lighting is high for an environmental monitoring application). Figure 4aand Figure 4b show the protection circuit for the power lines and the CAN bus lines,respectively. These circuits, which are separately applied to the power lines and theCAN bus since the latter is opto-isolated, share the same mechanism: the transientvoltage suppressors (i.e., G1 to G6) aim at preventing overvoltage peaks, while thechoke inductors (i.e., L1 to L6) handle overcurrent peaks. It is noted that we inserteda transient voltage suppressor for each couple of wires to be able to avoid overvoltagesamong VDD, circuit ground and case ground in the power lines and among the differ-ential CANH, CANL lines and case ground in the CAN bus line. Since the CAN bus is adifferential line, the choke inductors L3 to L6 are coupled to reduce noise. In addition,we introduced the protection electrostatic dischargers (PESD1 to PESD4) to protectcircuits from electrostatic discharges induced, for example, by human operators touch-


A:10

Vdd

G1

G2

G3D1D2

L1

L2

from

batteries

to

boards

Case Ground

GND

(a) Protection for power lines

CAN_H

G4

G5

G6D3D4

L3

L4

to the

board

Case Ground

L5

L6

CAN_L

from

bus

(b) Protection for CAN bus lines

Fig. 4. Lightning current/surge arrester

ing the circuits. The decision to design an ad-hoc lighting protection circuit (ratherthan considering off-the-shelf lighting protection circuits) is justified by the possibilityof integrating the circuit directly on the board (as depicted in Figure 2).

The dip switches circuit aims at storing the in-field configuration of the node (i.e.,the Node ID and CAN bus baud rate). The switches are manually fixed by designersand read, during the initialisation phase to configure the node, as detailed in Section 4of the companion paper. The dip switches circuit we chose is the Multicomp MCDHN-08-V with 8 switches and 1.27 pitch; other dip switch circuits with the same electroniccharacteristics could be considered as well.

Finally, the DC/DC power regulator circuit adapts the 12V provided through thepower supply line to the 3.3V requested by the processing and communication and thesignal conditioning boards. We adopted the TRACO TMR 1210, which operates in therange of 9-18VDC, with a conversion efficiency of 70% for an output voltage of 3.3V.

The advantage of a DC/DC regulator present in the processing and communicationboard of each SPU (instead of adopting just a centralised DC/DC at the BS) is twofold.Firstly, the galvanic isolation provided by the TRACO allows the ground of the cablefrom the one of the SPU boards to be decoupled, thus increasing the overvoltage pro-tection on the cable. In particular, the selected TRACO guarantees the managementof overvoltage up to 1000VDC. Secondly, the DC/DC at each SPU guarantees a sta-ble reference voltage for both the processing and communication board and the signalconditioning one. In particular, variations in the reference voltage of the signal condi-tioning board could induce errors in the analogue-to-digital conversion, thus affectingthe acquired measurements. Moreover, any glitches in the supply voltage of the micro-controller could generate spontaneous, unwanted reset actions.

4.2. The signal conditioning board

The signal conditioning board (see Figure 3c and 3d) contains the signal conditioningcircuits for the attached sensors. This board, together with the application code exe-cuted on the SPUs, is application-specific, whereas the processing and communicationboard (presented in the previous subsection) can manage a large class of applicationsbeing able to deal with both low and high bandwidth sensors.

As presented above and detailed in Section 4.3, the set of considered sensors encom-passes both low-bandwidth sensors, e.g., strain gauges, inclinometer and temperaturesensors, and high-bandwidth sensors, such as MEMS accelerometers and geophones.

Accordingly, the signal conditioning board hosts 6 conditioning channels for theMEMS sensor (3 for the tri-axial acceleration AC component and 3 for the tri-axialacceleration DC component), 3 channels for the geophones and one channel for thestrain gauge sensor.


A:11

MUX

A A

false

true

+20dB

1Hz

1Hz

A A

false

true

+20dB

sgampA

false

true

+20dB

+

Offsetsg

a)

b)

ax

mems

vx

geo

ggeo

gmems

gsmems

gsgeo

sx

tx

INsw

sgraw gssg

Fig. 5. Block scheme of the Signal conditioning board: a) Conditioning circuit for the x component of thegeophone and the MEMS accelerometer; b) Conditioning circuit for the strain gauge.

Figure 5 details the block scheme of the signal conditioning board. In particular,Figure 5a presents the signal conditioning circuit for the x channel of a geophone orMEMS accelerometer (channel y and z are omitted for figure readability), while Figure5b shows the signal conditioning circuit for the strain gauge. Geophone signals vxgeoare initially magnified by an amplification parameter of gain ggeo. Then, if necessary,signals can be further magnified by +20dB simply by setting the value of gsgeo to 1(logical TRUE).

Accelerometer signals axmems are initially filtered by means of analogue low-pass andhigh-pass filters with cut-off frequency at 1 Hz to separate the static component of thesignal tx (representing the projection of the gravitation acceleration ~g vector along thexth component) and the dynamic component (representing the acceleration along theaxes). The latter is then amplified by gmems with a value of up to +20dB (when gsmems

is equal to 1). The input INsw to the multiplexer specifies whether the geophone or theaccelerometer signals has to be considered; this input is set at the software level.

Strain gauge signal sgraw is amplified by +20dB if gssg is 1 and an offset Offsetsg isadded for compensation. In particular, the signal conditioning circuit for strain gaugeexposes a magnification ability which enables the remote operator to zoom a particularpart of the measuring range. The zooming factor has been fixed to 5, while the offseton the measuring range can be configured remotely.

The parameters of these circuits, summarised in Table II, have been designed to bemodified at run-time, thanks to digital potentiometers and digital switches (each SPUhas 24 tunable parameters). In particular, the digital potentiometers, connected to theDSPic through the SPI bus, adjust the impedance between the different operationalamplifiers comprising the analogue filters and amplifiers on the sensor board, whilethe digital switches, commanded through simple GPIO lines, alternate different setsof capacitors and resistors on the same circuits. This parametric-computation approachis a fundamental tool for designing robust adaptive monitoring systems able to adaptto changes in the environment.


A:12

Table II. SPU hardware parameters (dafault values are in bold)

Parameter Name Function Range

gmems MEMS signals amplifier gains (3 channels) (0− 40dB)gsmems MEMS gain switch (add 20dB to MG) {true, false}ggeo Geophone signals amplifier gains (3 channels) (8− 60dB)gsgeo Geophone gain switch (add 20dB to GG) {true, false}Offsetsg Strain gauge zoom offset (0− 255)gssg Strain gauge zoom enable {true, false}INsw Input selection switch {MEMS = 1, Geophone = 0}

4.3. Sensors

MEMS accelerometers and geophones are the sensors to be used for acquiring micro-acoustic signals in environmental or structural monitoring applications. To satisfy thisrequirement we considered a 2kHz sampling rate, hence MEMS accelerometers andgeophones are able to detect signals up to 1kHz in bandwidth. Such a bandwidth al-lows the SPUs to acquire signals in the band typical of both SHM applications (50Hzto 1kHz) and high sampling-frequency environmental applications (1Hz to 1kHz). Inaddition to a tri-axial accelerometer or geophone, each SPU mounts traditional sensorssuch as a strain gauge, an inclinometer and a temperature sensor.

The MEMS accelerometer used is the low power tri-axial ST LIS3L02AL developedby ST Microelectronics (range ±2g, resolution 0.5mg). We opted for an analogue ac-celerometer since off-the-shelf digital accelerometers provide lower sampling frequen-cies (sampling rates up to 250Hz). In addition, in our opinion, the selected accelerom-eter guarantees the best trade-off between performance and cost, while maintainingvery small sizes (5mm × 5mm). The three channels are sampled at 2kHz with a 12bitresolution. Similar to other MEMS accelerometers, the continuous component of theacquired acceleration measurements represents the projection of vector ~g on the axesof the MEMS, thus providing information about the inclination of the sensor.

The selected geophone is manufactured by Solgeo [Solgeo 2012] and includes a ternGeoSpace GS-20DH geophones to support three-dimensional sensing. These geophonesare passive and guarantee a bandwidth of up to 400Hz. Having said that, our electron-ics is able to host a large variety of geophones.

The temperature sensor used is the Microchip TC1047A. The temperature rangesfrom -40 °C to +125 °C with an accuracy of 0.5 °C and is sampled with a 12bit resolu-tion. Other temperature sensors providing the same electrical characteristics (supplyvoltage 3.3V, sensor output voltage 1.75V) would fit well with the electronics.

Among the wide range of strain gauges (with current output 4-20mA) available onthe market, we focused on the SISGEO 0D313SA1000 [Sisgeo 2012], for its sensitivity(operating range 300mm±50mm, resolution 0.01mm, repeatability < 0.3% of the fullscale). Also the strain gauge is sampled at 12 bits.

The sensors considered are summarised in Table III. It is worth noting that we can-not envisage (and it would be a nuisance) a continuous transmission of the acquireddata to the control room since microacoustic emissions are asynchronous burst-typeevents and the required data bandwidth is high (12bits · 3 channels · 2k samples).Moreover given the low expected number of events (sparse data over time) e.g., 0-10events per week, triggering mechanisms based on noise-to-signal ratio can be used atSPU level to detect the presence of micro-acoustic bursts [Alippi et al. 2007] and thedata is sent to the base station only when a possible event is detected.

4.4. The case

In monitoring systems designed to work in real-life (possibly harsh) environments, thecorrect design of the unit case is important as the correct design of the hardware and


A:13

Table III. SPU Sensors

Type Sensor model Range Resolution Accuracy

MEMS ST LIS3L02AL ±2g 0.5mg N.A.Geophone Three-dimensional board of

GeoSpace GS-20DH geophones28-400Hz na na

Strain gauge SISGEO 0D313SA1000 300mm ± 50mm 0.01mm < 0.3%FS

Temperature Microchip TC1047A −40°C to +125°C N.A. ±0.5

Fig. 6. The design sketch of the case.

software. Though several off-the-shelf cases or switch gears are available, we had todesign an ad-hoc case due to the peculiar application scenarios we are considering. Infact, traditional cases for WSN units are not designed to be hung from a rock face andcannot guarantee not to interfere with the signal.

Our unit case has been designed to be light and small in size. The final output ispresented in Figure 6 where the cylinder has a 120mm diameter and 100mm height.The considered material is aluminium: easy to produce, lighter than steel and barelyattenuates incoming microacoustic emissions. Moreover, the ability of aluminium toconduct heat allows the case to dissipate the internal thermal energy by convectionthrough the dissipating annular wings provided in the box case.

A prototype of the case was initially made and its performance tested under vibra-tions by means of a shaker (see Figure 8(a) and 8(b)). Unfortunately, the first prototypewas introducing a resonance at around 700Hz and therefore, interfering in the band-width of the signal.

We then modified the case by including grains in the upper part of the case to reducevibrations (these grains are highlighted in Figure 6). This allowed us to remove theresonance parasitic effect (as presented in Figure 8(c),8(d)).

It is noted that, as presented in Figure 7, the case has been designed in two maincomponents, i.e., the chassis and its support, to ease the deployment phase and main-tenance. Only the support is permanently fixed to the rock face (or the monitoringstructure) by means of holes and an ad-hoc resin. On the contrary, the chassis is an-chored to the support by means of screws. This solution simplifies maintenance; if aunit fails, we simply replace it with a new one.

5. THE BASE STATION AND THE CAN FIELD-BUS

The base station (BS) plays a crucial role in the hybrid wireless-wired architecturesince it powers the units, coordinates the communication activity among SPUs andserves as a gateway between the local system and the control room. Referring to Fig-


A:14

Cable inCable out

TM

External sensor

(strain gauge or

geophone)

Signal

Conditioning

Board

Processing and

Communication

Board Wire to board

connectors

Chassis

support

Chassis

Inter-board

Connectors

Rock

Substrate

Fig. 7. A section of the case.

ure 1, the BS acts as the master of the fieldbus by collecting the data acquired andstatus information (e.g., uptime, software version currently installed) from the SPUsand forwards them to the control room through the remote wireless channel. At thesame time, it provides mechanisms for time synchronisation at the SPUs level anddispatches commands back to the units. In its energy harvesting capacity, the BS ac-quires energy by means of photovoltaic cells based on a maximum power point tracker(MPPT) circuit and stores it into batteries (e.g., lead-acid batteries).

Figure 9 shows a picture of the base station whose block diagram representation isgiven in Figure 10.

In particular, the BS comprises

(1) the main board, which mounts the main microprocessor executing the applicationtasks;

(2) the energy harvesting board. It contains the MPPT and reports the status of theenergy harvester to the main board;

(3) the shut-down/wake-up board, which allows the main board to undergo duty cy-cling mechanisms;

(4) the energy management board, whose task is to enforce power management policiesdepending on the battery voltage;

(5) the radio module, either consisting of a radio link (e.g., a 5GHz WiFi) or an UMTSmodem.

It is noted that it is not required for the BS to be deployed on the rock face or on astructure close to the units. In fact, thanks to the fieldbus, it can be placed in a safeand easily reachable area. The BS is contained in the off-the-shelf Gewiss GW44811switch gear hung from a pole; the box size is 396 x 474 x 160mm.


A:15

(a) Response Module without grains (b) Response Phase without grains

(c) Response Module with grains (d) Response Phase with grains

Fig. 8. The Bode diagram (module and phase) of the case prototype without and with grains.

Fig. 9. A picture of the base station showing the main modules. The radio module is not here visible asintegrated in the antenna (which is outside the case).


A:16

Shut-down/

wake-up board

RTC

micro controller

Energy manager Board

MPPT_0 MPPT_1

Radio

Lin

k (

5G

hz W

iFi)

VD

DIN

Solar Panel

Batteries200Mhz ARM9

with 64Mbyte RAM

2GbyteSD card

ETH0

MPPT_2

Vdd

micro controller

to SPUs

ETH0

VDDIn

CAN

Solar Panel

Solar Panel

Energy harvesting BoardCOM0

COM1

COM2

to SPUs

COM3

Vddin

Main board

Vddin

GPIO

Fig. 10. The architecture of the base station

Similarly to SPUs, the main board is fitted with the lighting protection circuit (re-fer to Section 4) shielding both power and CAN bus lines. In addition, the BS is fit-ted with rechargeable batteries and solar panels to guarantee the system credibilityenergy-wise. We considered Lead-acid 12V 30Ah UNIBAT and the Lead-acid 12V 40AhUNIBAT batteries, which share the same technology but differ in the energy chargemechanism, size and weight. The choice between these two batteries is a design is-sue and is related to the amount of energy required by the system. We need to uselead-acid batteries since this technology guarantees a higher life-time of the batteryin terms of charge-discharge cycles. For solar panels, we considered a polycrystalline20W nominal solar panel manufactured by RALOSS (RALOSS-SR20-36). Other so-lar panels providing the same technical characteristics could have been considered aswell.

5.1. Main Board

The main board is also the mother-board for the whole system. Here, we opted forthe PC104-based Technology System TS7260 board based on an ARM9 200MHz mi-croprocessor and fitted with 64MByte of RAM, 14 general purpose I/O (GPIO) lines,four serial lines, a USB pen driver and an SD card reader. Moreover, the board is en-dowed with both a CAN and an Ethernet port to communicate with the SPUs andthe radio link, respectively. The GPIO communicates with the energy manager board,while the serial ports provide the communication link with the shut-down/wake-upand the energy harvesting boards. We chose the PC104 technology since it guaranteesboth lower power consumptions (w.r.t. other boards based on the x86 technology) andreduced size (which in our case is an important constraint). However, other PC104-based boards with similar electronics could have been considered as well. The mainboard stores both the operating system (Linux) and the application code in a read-only2GB SD card (Trascend Industrial Grade TS2GSD80I 2GB SD card) to prevent anyunintentional modification of the code, while data (collected from SPUs and ready to


A:17

Solar

Panel

DC/DC

regulator

Control CPU

Battery

Energy status

information

(UART)

Vs

Ic

Vc

Ip

Fig. 11. The block diagram of MPPT circuits. Vs, Ip, Vc and Ic are the voltage of the solar panel, the currentgenerated by the solar panel, the output voltage and current from the DC/DC regulator, respectively.

be transmitted to the Control Room) are stored as files to the USB pen drive (STECIndustrial Grade SLUFD2GU1U USB 2GB Flash Drive). In case of a USB pen drivefault, this solution allows the BS to inform the control room about the fault occurred.It is noted that both the SD card and the USB pen drive must be certified as industrialgrade so as to work in extreme temperatures ranging between -40°C to +85°C interval.

The application code, detailed in [Alippi et al. 2012], consists of a script polling theSPUs for data acquisition, storing data to files, opening the connection with the re-mote control room and transmitting data. Finally, it receives commands that, suitablyparsed, are dispatched back to the targeted SPU. The CAN port of the main boardallows for the BS to be connected with the SPUs by means of the industrial CANbus(version 2.0b) providing both the physical and the MAC layers in hardware. The max-imum bandwidth is 1Mbps for cables up to 40m and decreases with the cable length(e.g., the maximum bandwidth for a 100m long cable is 250kbps).

Although several transport layers for CANbus have been presented in the literature(e.g., see CANOpen [Lawrenz 1997]), they cannot be used in energy aware applica-tions since the energy aspect is not tackled: being designed for industrial applications,CANOpen requires the units to be always on. The proposed transport layer designedto tackle this issue will be discussed in the companion paper [Alippi et al. 2012]. TwoGPIO lines provided on the main board are particularly interesting and allow the SPUsto be switched on and off at once or enable the remote radio link. These mechanismsguarantee implementation of fine-grain energy management policies.

5.2. The energy harvesting board

As stated in [Alippi and Galperti 2008], solar panels exhibit a non-linear electricalcharacteristic which is a function of the solar radiation and the operational tempera-ture and are affected by ageing, dust and efficiency degradation. These time-varyingnon-linear effects can make traditional (diode-based) energy harvesting mechanismsvery inefficient as they require the solar panel voltage to be basically above that of thestorage means.

The suggested energy harvesting board relies on a maximum power point tracker(MPPT) circuit [Alippi and Galperti 2008] which sets the voltage of the solar panel toan optimal value, maximising the power transferred. This optimisation step is carriedout by continuously adapting the working point of the photovoltaic cells to the optimalvalue. In addition, the MPPT circuit used, presented in Figure 11, consists of twomain blocks: an adaptive digital control loop and a voltage input controllable DC/DCconverter stage. The first block consists of a microcontroller, whose code adaptivelyidentifies the optimal working point for the solar cells by setting the optimal voltageVs and maximising the extracted power. In addition, the microcontroller reports thebattery voltage status to the main board through a serial port for subsequent decisionmaking. The second block, which is implemented by means of a single-ended primary-


A:18

Normal

Mode

Warning

Mode

Safe

Mode

Vbatt ≥ 12V

Vbatt ≥ 12V

Vbatt < 8V

Vbatt < 9V

Vbatt < 8V

Fig. 12. Behaviour of the energy manager

inductor converter (SEPIC) circuit, aims at adapting the input (cells) voltage identifiedby the microcontroller of the solar panel with that of the storage battery.

The energy harvesting board presented in Figure 10 consists of three indepen-dent MPPT circuits, each one controlling a single solar panel. Nevertheless, as de-scribed in Section 7, the number of solar panels (and consequently of MPPT circuits)is application-specific and depends on the power consumption, duty-cycle of the unitsand expected solar radiation of the deployment area. We decided to develop an ad-hocenergy harvesting board since off-the-shelf energy harvesting boards (e.g., traditionalsolar charge regulators) are not tailored for small harvested energies (hence, the en-ergy consumption of these boards is not negligible for the energy consumptions of oursystem). In addition, the energy harvesting board developed provides energy informa-tion about the system such as the panel voltage, generated current and battery voltage.Such information, as presented in Section 5.3, is essential for an effective energy man-agement. The MPPT circuit has been proved to be particularly effective even in cloudyor rainy days, situations where traditional diode-based energy harvesters proved to beinefficient at low power solar radiations. In fact, the traditional diode-based solutionsare not able to harvest power when the incident solar power is below 10mW, while theused MPPT circuit can operate with incident solar powers down to 2mW, a commonsituation in foggy, misty or rainy days [Alippi and Galperti 2008].

5.3. The energy management board

The aim of the energy management board, whose block diagram is detailed in Figure10, prevents the battery voltage to drop below a fixed threshold (i.e., 8V). We decided todevelop an ad-hoc energy management board since the battery management circuitsavailable on the market (e.g., traditional solar charge regulators) do not provide ad-vanced energy management policies. Our energy management board implements anadvanced energy management policy as summarised in Figure 12. In particular, whenthe battery voltage is above 9V (normal mode), the system is fully operational and nopolicies must be undertaken. When the battery voltage is between 8V and 9V (warningmode), the energy management board cuts off the power supply to the SPUs and theradio module, waits till the end of the duty-cycle of the main board and shuts down.When this happens, the energy management board idles until the voltage battery isback to 12V (thanks to solar recharging) to switch back to normal mode. We introduceda safety threshold at 8V (safe mode) to address unpredictable cases where the mainboard does not complete the shut-down operation (e.g., due to software errors, kernelpanic), thus further consuming the residual energy. When the battery level drops be-low 8V, the energy management board cuts off the power supply of the main boardregardless of shut-down completion. Even in this case, the energy management board


A:19

restores the normal operational mode only when the battery voltage returns above12V.

5.4. The shut-down/wake-up board

The shut-down/wake-up board, whose block diagram is presented in Figure 10, allowsthe main board to undergo duty-cycling by switching the BS off for a certain amountof time and waking it up again when needed. The board used is the off-the-shelf Tech-nology Systems TS-BAT3, which is a PC104 board and can be easily integrated withthe PC104 main board. The selection of the board was based on compatibility issues.The main board periodically prompts the shut-down/wake-up board to shut down fora certain amount of time to save energy; this action is carried out through a messagesent along the serial connection to the board. As a reply, the shut-down/wake-up boardactivates a timer (for the amount of time requested by the main board) and activates arelay that cuts-off the power line of the main board. Then, when the timer expires, theshut-down/wake-up board re-enables the relay, thus powering the main board, whichreboots. The shut-down/wake-up board consists of a low-power micro-controller, a re-lay, a Real Time Clock (RTC) powered by a battery button, and a serial port connectedto the main board. In particular, the micro-controller remains in a deep-sleep state un-til it receives an interrupt on the serial port. Thereafter, it receives the messages fromthe main board (containing sleeping time information) and, finally, activates an RTCtimer. Afterwards, the micro-controller cuts off the power of the main board by meansof the relay (that switches according to a GPIO line of the micro-controller) and entersinto a deep-sleep state again to further reduce energy consumption. When the timerexpires, the RTC generates an interrupt which wakes up the micro-controller whichthen acts on the relay to restore energy back.

5.5. The radio module

The radio module provides the remote radio link by connecting the BS with the remotecontrol room. Two main technologies, which differ by bandwidth, power consumptionand coverage area, have been considered: a radio link and a 3G UMTS modem. Morespecifically, the 5GHz WiFi radio link guarantees a 6Mbps TCP/IP channel for dis-tances up to 7Km, with a peak power consumption of 4W. The main drawback of thissolution is the line-of-sight requested between the antennas of the RMS (directive)and that of the remote control room. The WiFi link we selected is the Fly Communica-tions Nebula 2458. Other 5GHz WiFi links available on the market could be consideredas well. On the contrary, the UMTS modem relies on the coverage of the UMTS mo-bile network (no line-of-sight required) at the expenses of higher power consumption(up to 18W) and a lower bandwidth (in theory up to 5.76Mbps in up-link; in practice,generally from 500kbps to 1.5Mbps). When the UMTS network is not available, themodem automatically switches to a GPRS connection (max 14.4kbps in up-link), in-creases the duty-cycle (see Section 7) and accordingly, power consumption. We chosethe Quad-band DIGI Connect WAN 3G modem for its flexibility (thanks to its quad-band technology) and since it can be easily controlled through the Ethernet port.

6. DESIGN OF THE SYSTEM FROM THE ENERGY POINT OF VIEW

As described in Section 4 and 5, the design of both the SPUs and the BS was low power-oriented with particular attention to the choice of the components and the hardwaredesign. In addition, for the system to operate for years in harsh working conditions,a correct design of batteries and solar panels must be carried out. An incorrect ratingof the energy harvesting module for the system can led to two unpleasant effects: 1)if the system is under-rated w.r.t batteries and solar panels, the system will run outof energy frequently (with a consequent frequent loss of measurements); 2) an over-


A:20

rated system increases the complexity, size and cost of the system. The correct designof the energy harvesting and power supply modules certainly depends on the systemenergy consumption which is in turn a function of the power consumption of its com-ponents and duty cycle. Let’s assume that one BS and N SPUs (the number of SPUs isapplication-dependent and defined at the design stage) are used. The power consump-tion (measured in laboratory tests) of the main functional elements is given in TableIV.

Table IV. Power consumption (in W) for one SPU, the BS and thelong-range radio link

PSPU PBS PWiFi PGPRS

Power Consumption (W) 0.3 2.4 4 18

Let τSPU , τBS , τWiFi, τGPRS be the duty-cycles of each SPU, BS, WiFi and GPRSlink, respectively (SPUs are always operational and cannot be switched off, i.e.,τSPU = 100%). The BS is fitted with either the WiFi or the GPRS link dependingon the deployment (the duty cycle of the link not present in the BS is 0).

The values of PSPU , PBS , PWiFi and PGPRS , together with τSPU , τBS , τWiFi, τGPRS

and N allow us to evaluate the energy consumption of the whole system in each cycle(fixed at 60 minutes for both deployments, yet this is a parameter that can be remotelychanged):

Esys = 3600 ·(

N · PSPU · τSPU + PBS · τBS + P{WiFi/GPRS} · τ{WiFi/GPRS}

)

(1)

The energy Ebatt (in J) stored by a new lead-acid battery can be roughly estimatedas:

Ebatt = CHnm · Vnm · 1hour (2)

where CHnm and Vnm are the electric charges transferred by the battery in one hourand the nominal voltage of the battery, respectively. Thus, to identify the minimumnumber of batteries able to guarantee at least hmin hours without any solar energyharvesting, we must satisfy the bound:

Nbatt ·Ebatt

Esys> hmin (3)

where Nbatt is the number of batteries, in a parallel configuration. The selection ofthe number of solar panels for each deployment requires information about the solarradiation in the deployment area. We run a 3 months acquisition campaign of solarradiation information in winter by means of two ad-hoc embedded systems, each fittedwith a nominal 20W solar panel. Let T be the time horizon of the acquisition campaignand let’s consider an energy harvesting circuitry as the one presented in Figure 11, theenergy (in J) acquired by one solar panel is:

Esp =

∫

T

Vs(t)Ic(t)dt (4)

where Vs(t) and Ic(t) are the voltage of solar panel and the current from the DC/DCregulator at time t, respectively. Let ET

sys be the energy consumption of the system inthe time horizon T ,


A:21

ETsys =

∫

T

(

N · PSPU · τSPU + PBS · τBS + P{WiFi/GPRS} · τ{WiFi/GPRS}

)

dt. (5)

Thus, to identify the minimum number of solar panels able to guarantee the re-quested long-term energy sustainability for the system we need:

Nsp · Esp > ETsys (6)

where Nsp is the number of solar panels. An energy-wise effective system design re-quires the definition of Nbatt and Nsp.

7. EXPERIMENTAL RESULTS

The rock collapse forecasting application is presented in Subsection 7.1, while currentdeployments are given in Subsection 7.2; Subsection 7.3 describes the system designenergy-wise. Finally, Subsection 7.4 introduces and discusses the system performancein terms of energy-efficiency, ability to deal with high sampling-rates and Quality-of-Service (QoS).

7.1. Real-world application: a monitoring system for rock collapse forecasting

Rocks collapse is one of the most dangerous and sudden natural hazards in mountainregions. This harmful event, which is generally characterised by catastrophic conse-quences if human settlements, transport routes or critical infrastructure systems areinvolved, is related to the fall of rocks or stones induced by the (partial) collapse of arock face due to frost-defrost phenomenon, thermal stress and gravitation.

The forecasting of rock collapse nowadays represents a challenging research issueand a valuable testbed for the high frequency monitoring system proposed for sev-eral reasons. Firstly, the collapse of a rock face is critical due to the lack of clear andeffective forerunners. Secondly, traditional investigation techniques (e.g., those basedon strain gauge and inclinometer sensors), mainly relying on data loggers, cannot beused for real-time decision making. Thirdly, the possibility to forecast a collapse in arock is related to the availability of internal information associated with micro acous-tic/seismic activities induced, at the microscopic level, by the creation and organisationof fractures from small to large dimensions.

MEMS accelerometers and geophones are the right sensors to be used for acquir-ing micro-acoustic signals (a micro acoustic emission is similar to a small earthquakegenerated within the rock face). Herein, we considered a 2kHz sampling rate since thedesigned system was able to support such a frequency, thus being able to detect sig-nals up to 1kHz in bandwidth. A high synchronism among the acquisition units, notexceeding 1ms in skew, is required to carry out the localisation of the emission source.In addition to a 3 axis accelerometer or geophone, each SPU mounts traditional sen-sors such as strain gauge, inclinometers and temperature sensors. The SPU unit ishosted in the metallic case described in Section 4.4 and shown in Figure 13.

7.2. Current deployments: the St. Martin mount and the Rialba’s tower

Currently, we have had two monitoring systems operational for more than two years onthe Alps, in northern Italy: the first one on the St. Martin mount, the second one on theRialba’s towers (both on the Lake Como area). Another deployment has been recentlydone in Switzerland. For all deployments, geologists and geophysicists investigated theareas to identify the number and location for the SPUs. Moreover, due to the presenceof metallic units, cables, and antennas, we positioned lightning rods close to the unitsto complement and integrate the internal lightning protections.


A:22

Fig. 13. A mounted SPU case.

(a) St. Martin mount (b) Rialba’s towers

Fig. 14. The deployment areas at the St Martin mount and Rialbas towers

The St. Martin mount is a 300m high rockface near the city of Lecco, which expe-rienced a 15000m3 rock collapse in 1969 causing 7 fatalities. The monitoring system,which consists of eight SPUs with a different sensor platform (6 MEMS accelerome-ters, 2 geophones, 2 strain gauges, 8 temperature sensors, 8 MEMS inclinometer sen-sors) and a BS, has been active since April 1st, 2010. The BS is equipped with a 5GHzWiFi radio link with a directional antenna oriented towards our Intelligent EmbeddedSystem lab (which is located at Lecco’s Campus of Politecnico di Milano) at a distanceof approximately 2.5Km. The total length of the wired segments accounts for 100mconnected with a CanBUS modality set to 250Kbs.

The second deployment area is the Rialba’s towers, a limestone-conglomerate rockdivided by a system of fractures. The towers are close to the main civil infrastructuresconnecting Lecco and Sondrio provinces (speedway, railway, gas and electric distribu-tion networks). The monitoring system consists of three SPUs and a BS and has beenactive since July 7th, 2010. The base station relies on a UMTS modem for remote dataand commands communication with our Intelligent Embedded System lab.


A:23

(a) A SPU with a geophone (b) A SPU with a strain gauge

Fig. 15. Current deployments.

7.3. Design of the system from the energy viewpoint

To correctly dimension each deployment energy-wise, in terms of number of batteriesand solar panels, we used the methodology presented in Section 6 by taking into ac-count the different number of units per system (which is application-dependent) andthe different technologies considered for the long-range communication (i.e., WiFi forthe system at the St. Martin mount and GPRS for the one at Rialba’s towers). Thepower consumption of each SPU PSPU , of the BS PBS and the long-range radio linkPR for both deployments are given in Table IV; Table V shows the duty-cycles of SPUsτSPU , BS τBS and long-range radio link τR. It is noted that SPUs must be always op-erational (continuous data acquisition and processing) and cannot be switched off (i.e.,τSPU = 100%).

Table V. Duty-cycle (%) for SPU, BS and long-rangeradio link and number of SPUs in the deployments

Deployment τSPU τBS τR N

St. Martin mount 100 16.6 1.6 8Rialba’s towers 100 16.6 3.6 3

The values of PSPU , PBS and PR, together with τSPU , τBS , τR and N allow us toevaluate the energy consumption of the whole system in each cycle (fixed at 60 minutesfor both deployments, but this parameter may be changed remotely) by using Eq. 1.These energy consumptions are summarised in Table VI; Figure 16 shows the time-profile of the power consumption per cycle for both deployments.

Table VI. Energy consumption (kJ) in a cycle (60mins)

Deployment Energy consumption (kJ)St. Martin mount 10.3Rialba’s towers 7.0

We dimensioned both deployments to survive at least 10 days without any energyharvesting (i.e., hmin = 240 hours). By using Eq. 3 and considering a standard lead-acid battery (e.g., Vnm = 12V , CHnm = 30Ah), we obtained Nbatt = 2 batteries for bothsystems. It is worth noting that with such dimensioning, we had the system surviveto 15 days of continuous rain thanks to the MPPT energy harvester, which was able to


A:24

10 20 30 40 50 600

5

10

15

20

25

Time (min)

Pow

er(W

)

Power consumption per cycle

Towers of Rialba

St. Martin mount

Fig. 16. Time-profile of the power consumption per cycle (60 mins) for both deployments.

extract up to 3W (for a 20W nominal panel) even on cloudy days (e.g., see ’bad weather’in Figure 18).

The selection of the number of solar panels for each deployment required a detaileddesign phase. As described in Section 6, we started with a long-term acquisition cam-paign of solar radiation information lasting 3 months (i.e., T = 3 months) in winter bymeans of two ad-hoc embedded systems fitted with a nominal 20W solar panel. Thesedata, together with a detailed model of both the energy harvesting circuitry and thepower consumption of the systems considered, allowed us to select Nsp = 3 nominal20W solar panels for the system at the St. Martin mount and Nsp = 2 nominal 20Wsolar panels for the Rialba’s towers.

7.4. Measurements from the deployments

We provide the real data to show the energy sustainability of the systems (by lookingat solar-generated power and the battery voltage), the capability to sample up to 2kHz(by looking at an acquired burst both in time and frequency domain) and the QoS (bylooking at the uptimes of the system units). In more detail, the solar generated powerand the voltage of the batteries for the July 2010-July 2011 period are given in Figure17 for the Rialba’s towers system. The data, which refer to one solar panel and onebattery, show the system capability to keep the battery voltage always above 12V forthe entire one-year period.

Figure 18 shows a detail of Figure 17 by presenting the harvested power and thebattery voltage between December 11, 2010 and January 13, 2011. This figure is par-ticularly interesting since the period considered includes two time slots of consecutivebad weather conditions. In more detail, from December 21 to December 27 the solarpanels did not acquire any energy due to heavy rainfalls. The battery voltage decreasedfrom 13V to approximately 12V but the system was able to keep all of its functional-ities active (i.e., acquisition, processing and remote transmission). Afterwards, somesunny days allowed the battery voltage to recover. Likewise, the period between Jan-uary, 4th and January, 13th was characterised by bad weather conditions causing thebattery voltage to drop again. The acquisition measurements of January, 3rd 2011 areinteresting and show the capability of the MPPT board (as stated in Section 5.2) to har-


A:25

Fig. 17. The generated solar power and the voltage of the battery for the period July 2010- July 2011,Rialba’s towers : a) generated solar power b) battery voltage

Rainy days

Cloudy day

Fig. 18. A detail of the generated solar power and the voltage of the battery for the period December 11th,2010- January 13th, 2011, Rialba’s towers: a) generated solar power b) battery voltage.

vest even small quantity of energy when available (see the small peaks in the batteryvoltage).

Figure 19 presents the average solar generated power per month for both the solarpanels in the July 2010-June 2011 period. In 2011 in northern Italy we experienced an


A:26

Fig. 19. Average acquired solar generated power per month for the period July 2010- June 2011 .

exceptionally sunny and hot April and a very bad June, which is well reflected by thevery high value of generated power in April 2011 compared to that of May and June.

We compared the average solar generated power acquired per month to the valueprovided by ENEA (Italian National agency for new technologies, Energy and sustain-able economic development) in the same area of the deployments during the years1994-1999 [ENEA 1999]. To compute these values we considered a solar panel as be-ing the same size as the ones we considered and a panel efficiency equal to 0.3. Theexpected results are in line with the generated power acquired by our system. Thedifferences between the expected and the measured solar power could be ascribed tothe specific position and orientation of the solar panel considered and to the fact thatour measurements are related to a single year (while ENEA data are averaged overfive years). Having this in mind, data from ENEA can be safely used to dimension anyother deployment needs without carrying out a preliminary, in-situ investigation.

Figure 20 shows an example of a burst detected by the system at the Rialba’s towersduring its monitoring activity. In particular, Figures Figures 20(a), 20(c) and 20(e)detail the three accelerometer output components (X, Y and Z) of the acquired burst. Asexpected by geologists and geophysicists, the analysis in the Fourier domain, presentedin Figures 20(b), 20(d) and 20(f), shows that the signal frequencies of the bursts rangefrom 400Hz to 900Hz. A more detailed analysis of the acquired burst is presented inthe companion paper [Alippi et al. 2012].

To evaluate the QoS of the proposed system we considered the uptime counter for theSPUs and the BSs in both deployments (these data logs are sent together with senseddata to the control room). Such analysis allows us to identify possible long-term, nondevastating bugs not recognisable through laboratory tests.

To measure the uptime counter of a BS, we rely on a counter of the duty cycles (atimestamp is associated with each value of the counter). Since the duty-cycling periodis regular (e.g., five minutes), the corresponding plot is a line (whose slope is directlyaffected by the duty cycle). Figure 21 shows the counters from the BSs of both thedeployments. In the St. Martin mount deployment we can observe an irregular be-haviour: the counter reset three times in the first six months and four times in thefirst year of the deployment.

The reason is due to a corruption of the file system caused by a software bug inthe code running on the BS and blocking the system. We had to reformat the USBpen drive to reactivate the system. In June 2011, we modified the duty cycle of the


A:27

0 20 40 60 80 100 120 140-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

Time (ms)

Accele

ration (

g)

(a) Acceleration axis X

0 200 400 600 800 10000

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Frequency (Hz)

Single-Sided Amplitude Spectrum

(b) FFT of the axis X

0 20 40 60 80 100 120 140-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

0.06

0.08

Time (ms)

Accele

ration (

g)

(c) Acceleration axis Y

0 200 400 600 800 10000

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Frequency (Hz)


(d) FFT of the axis Y

0 20 40 60 80 100 120 140-0.06

-0.04

-0.02

0

0.02

0.04

0.06

Time (ms)

Accele

ration (

g)

(e) Acceleration axis Z

0 200 400 600 800 10000

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Frequency (Hz)


(f) FFT of the axis Z

Fig. 20. An example of a detected burst (X,Y and Z components) and its Fourier plot


A:28

Fig. 21. BS uptime counter over time

Table VII. Deployment QOS

Deployment Deployment duration(days) QOS(%)St. Martin mount 534 84.1Rialba’s towers 471 96.7

BS at the St. Martin’s deployment to 2.5 minutes, as explained by the change in theline slope. The Rialba’s towers system is an evolution of the system deployed at theSt. Martin mount where the software bug causing the file system problem has beenfixed. This resulted in a significant improvement of the QoS. In particular, Figure 21demonstrates shows that the reliability of the Rialba’s towers system largely improvedcompared to that at the St. Martin mount since the counter never reset. In Table VII,we summarise the QoS for the two systems in terms of total uptime counter w.r.t. thedeployment duration up to October 2011. It is worth noting that both the system at theSt. Martin mount and the Rialba’s towers underwent a data loss in the counter plot.This was due to a software bug on the server software of the control room: in particular,an unhandled exception caused the abnormal termination of the thread responsible forsaving the gathered data to the database, causing the data loss.

In addition to the QoS of the BSs, we evaluated the uptime counters of the SPUs totake into account possible soft-resets (e.g., caused by the watch dog timer) or communi-cation timeouts (e.g., due to communication problems). To this end, we rely on two 16-bit counters provided by the FreeRTOS operating system, named tick and overflows.The first one is incremented every OS ticks, the latter when the counter tick overflows.Then, the corresponding uptime counter is computed as overflows ∗ 216 + (tick/500).When both overflows and tick counters overflow, the corresponding uptime counterrestarts at zero. Moreover, these counters reset to zero after every node reset becausethey are stored to the RAM. In particular, Figure 22 presents the uptime counters ofNode 1 and 4 deployed in St. Martin mount (the other units provide similar resultsand their uptime counters are omitted for brevity), while Figure 23 shows the uptimecounter for Rialba’s towers units. It is noted that in all the plots, the uptime counter


A:29

Fig. 22. St. Martin mount: uptime counters of Nodes 1 and 4.

Fig. 23. Rialba’s towers: uptime counters of deployed nodes.

overflows frequently, meaning that no soft-reset or communication timeouts occurred.In particular, the deployment at Rialba’s towers is not affected by these types of faults(the corresponding uptime counter always reaches its maximum value). On the con-trary, the SPUs at the St. Martin mount experienced resets induced by problems at theBSs (due to the randomly activating software bugs). In addition, in February and June2011 the counters of all the SPUs at the St. Martin mount reset to zero for maintenanceactions on the BSs. Table VIII shows the QoS for each node of the two deployments.

8. CONCLUSIONS

The hardware design of a hybrid wireless-wired monitoring system for high-frequencysampling has been presented in this paper. Unlike the solutions found in the literature,the chosen architecture and the ad-hoc designed electronics guarantee both long-termsystem availability (by means of energy efficient policies and energy harvesting mech-anisms) and flexibility and adaptability to changes in the network, environment or ap-plication requirements (through a remote reconfigurability of the units). The effective-ness of the proposed monitoring system has been tested in a challenging monitoringapplication for the forecasting of rock collapse, an application characterised by strictconstraints such as high sampling rate (up to 2kHz), high synchronisation among theacquisition units (down to 0.5 ms) and difficulties in the energy harvesting. Two de-ployments have been presented in this paper, which have been active in remote harshenvironments on the Alps for almost two years.


A:30

Table VIII. Per node QOS

Net Node QOS(%)

St.

Mart

inm

oun

t 1 84,52 82,93 82,94 83,15 84,86 84,87 84,88 84,8

Ria

lba’s

tow

er

1 97.02 97.03 96.1

ACKNOWLEDGMENTS

The authors would like to sincerely thank Prof. Gourab Sen Gupta, Massey University, New Zealand, forthe precious help in improving the quality of the manuscript.

REFERENCES

AKYILDIZ, I., SU, W., SANKARASUBRAMANIAM, Y., AND CAYIRCI, E. 2002. A survey on sensor networks.Communications Magazine, IEEE 40, 8, 102–114.

ALIPPI, C., BORACCHI, G., CAMPLANI, R., MARULLO, A., AND ROVERI, M. 2012. An hybrid monitoringsystem for the real-time processing of high frequency data. part a: Software aspects. Tech. rep. InternalReport.

ALIPPI, C., CAMPLANI, R., AND GALPERTI, C. 2007. Lossless compression techniques in wireless sensor net-works: Monitoring microacoustic emissions. In Robotic and Sensors Environments, 2007. ROSE 2007.International Workshop on. IEEE, 1–5.

ALIPPI, C., CAMPLANI, R., GALPERTI, C., MARULLO, A., AND ROVERI, M. 2010. An hybrid wireless-wiredmonitoring system for real-time rock collapse forecasting. In Mobile Adhoc and Sensor Systems (MASS),2010 IEEE 7th International Conference on. IEEE, 224–231.

ALIPPI, C. AND GALPERTI, C. 2008. An adaptive system for optimal solar energy harvesting in wirelesssensor network nodes. IEEE Transactions on Circuits and Systems I: Regular Papers 55, 6, 1742–1750.

BASHARAT, A., CATBAS, N., AND SHAH, M. 2005. A framework for intelligent sensor network with videocamera for structural health monitoring of bridges. In Pervasive Computing and Communications Work-shops, 2005. PerCom 2005 Workshops. Third IEEE International Conference on. IEEE, 385–389.

BOLLER, C., CHANG, F., AND FUJINO, Y. 2009. Encyclopedia of Structural Health Monitoring. Wiley.

CAICEDO, J., MARULANDA, J., THOMSON, P., AND DYKE, S. 2002. Monitoring of bridges to detect changesin structural health. In American Control Conference, 2001. Proceedings of the 2001. Vol. 1. IEEE, 453–458.

CROSSBOW. 2009. http://www.xbow.com/.

DELL’ACQUA, A., HANSEN, M., IKINEN, S., LOFSTEDT, B., VANUXEM, J., SVENSSON, C., YUAN, J.,HENTZELL, H., DEL BUONO, L., DAVID, J., GENAT, J., LEBBOLO, H., LEDORTZ, O., NAYMAN, P.,SAVOY-NAVARRO, A., ZITOUN, R., ALIPPI, C., BREVEGLIERI, L., DADDA, L., PIURI, V., SALICE, F.,SAMI, M., STEFANELLI, R., CATTANEO, P., FUMAGALLI, G., GOGGI, G., BRIGATI, S., GATTI, U., MAL-OBERTI, F., TORELLI, G., CARLSON, P., KEREK, A., APPELQUIST, G., BERGLUND, S., BOHM, C., EN-GSTROM, M., YAMDAGNI, N., SUNDBLAD, R., HOGLUND, I., AND PERSSON, S. 1993. System level poli-cies for fault tolerance issues in the fermi project. In Defect and Fault Tolerance in VLSI Systems, 1993.,The IEEE International Workshop on. 1 –8.

EISENMAN, S., MILUZZO, E., LANE, N., PETERSON, R., AHN, G., AND CAMPBELL, A. 2009. Bikenet: A mo-bile sensing system for cyclist experience mapping. ACM Transactions on Sensor Networks (TOSN) 6, 1,6.

ENEA. 1999. http://clisun.casaccia.enea.it/pagine/tabelleradiazione.htm. In Tabella della Radiazione So-lare, ENEA - Italian National agency for new technologies, Energy and sustainable economic develop-ment.


A:31

ENGEL, J., ZHAO, L., FAN, Z., CHEN, J., AND LIU, C. 2004. Smart brick-a low cost, modular wireless sensorfor civil structure monitoring. In International Conference on Computing, Communications and ControlTechnologies (CCCT 2004), Austin, TX USA.

FARRAR, C., PARK, G., AND TODD, M. 2011. Sensing network paradigms for structural health monitoring.New Developments in Sensing Technology for Structural Health Monitoring, 137–157.

FARRAR, C. AND WORDEN, K. 2007. An introduction to structural health monitoring. Philosophical Trans-actions of the Royal Society A: Mathematical, Physical and Engineering Sciences 365, 1851, 303.

GLASER, S. 2005. Terra-Scope: a MEMS-based vertical seismic array. Proc. SPIE.

GRATTAN, S., TAYLOR, S., BASHEER, P., SUN, T., AND GRATTAN, K. 2009. Structural health monitoring-better solutions using fiber optic sensors? In Sensors, 2009 IEEE. IEEE, 811–814.

GUNGOR, V. AND HANCKE, G. 2009. Industrial wireless sensor networks: Challenges, design principles,and technical approaches. Industrial Electronics, IEEE Transactions on 56, 10, 4258–4265.

HU, W., BULUSU, N., CHOU, C., JHA, S., TAYLOR, A., AND TRAN, V. 2009. Design and evaluation of a hybridsensor network for cane toad monitoring. ACM Transactions on Sensor Networks (TOSN) 5, 1, 4.

INGELREST, F., BARRENETXEA, G., SCHAEFER, G., VETTERLI, M., COUACH, O., AND PARLANGE, M. 2010.Sensorscope: Application-specific sensor network for environmental monitoring. ACM Transactions onSensor Networks (TOSN) 6, 2, 17.

JOHNSON, W., AULD, B., AND ALERS, G. 1994. Spectroscopy of resonant torsional modes in cylindrical rodsusing electromagnetic-acoustic transduction. The Journal of the Acoustical Society of America 95, 1413.

KIM, S., PAKZAD, S., CULLER, D., DEMMEL, J., FENVES, G., GLASER, S., AND TURON, M. 2007. Healthmonitoring of civil infrastructures using wireless sensor networks. In Proceedings of the 6th interna-tional conference on Information processing in sensor networks. ACM, 254–263.

LAWRENZ, W. 1997. Can System Engineering: From Theory to Practical Applications 1st Ed. Springer-VerlagNew York, Inc., Secaucus, NJ, USA.

LIM, S. AND SHOURESHI, R. 2003. Neural-based monitoring system for health assessment of electric trans-mission lines. In American Control Conference, 2003. Proceedings of the 2003. Vol. 3. IEEE, 2270–2275.

LIM, S. AND SHOURESHI, R. 2006. Advanced monitoring system for integrity assessment of electric powertransmission lines. In American Control Conference, 2006. IEEE, 6.

LYNCH, J., LAW, K., KIREMIDJIAN, A., CARRYER, E., FARRAR, C., SOHN, H., ALLEN, D., NADLER, B., AND

WAIT, J. 2004. Design and performance validation of a wireless sensing unit for structural monitoringapplications. Structural Engineering and Mechanics 17, 3-4, 393–408.

LYNCH, J., PARTRIDGE, A., LAW, K., KENNY, T., KIREMIDJIAN, A., AND CARRYER, E. 2003. Design ofpiezoresistive MEMS-based accelerometer for integration with wireless sensing unit for structural mon-itoring. Journal of Aerospace Engineering 16, 108.

MAROTI, M., KUSY, B., SIMON, G., AND LEDECZI, A. 2004. The flooding time synchronization protocol. InProceedings of the 2nd international conference on Embedded networked sensor systems. ACM, 39–49.

MOTEIV. 2007. Tmote invent. http://www.moteiv.com/.

PAEK, J., CHINTALAPUDI, K., CAFFREY, J., GOVINDAN, R., AND MASRI, S. 2005. A wireless sensor networkfor structural health monitoring: Performance and experience.

PAEK, J., GREENSTEIN, B., GNAWALI, O., JANG, K., JOKI, A., VIEIRA, M., HICKS, J., ESTRIN, D., GOVIN-DAN, R., AND KOHLER, E. 2010. The tenet architecture for tiered sensor networks. ACM Transactionson Sensor Networks (TOSN) 6, 4, 34.

RAMAMURTHY, H., PRABHU, B., GADH, R., AND MADNI, A. 2007. Wireless industrial monitoring and con-trol using a smart sensor platform. IEEE Sensors Journal 7, 5, 611–618.

RICE, J. AND SPENCER JR, B. 2008. Structural health monitoring sensor development for the Imote2 plat-form. In Proceedings of the SPIEThe International Society for Optical Engineering. Vol. 6932. 693234–1.

RICE, J. AND SPENCER JR, B. 2009. Flexible smart sensor framework for autonomous full-scale structuralhealth monitoring. Newmark Structural Engineering Laboratory Report Series 018.

SCHMID, J., G\\”ADEKE, T., STORK, W., HENNRICH, H., AND BLANK, T. 2010. A wireless MEMS-sensornetwork concept for the condition monitoring of ball screw drives in industrial plants. In Proceedings ofthe 8th ACM Conference on Embedded Networked Sensor Systems. ACM, 425–426.

SISGEO. 2012. Home page. http://www.sisgeo.com/.

SOLGEO. 2012. Home page. http://www.solgeo.eu/.

WERNER-ALLEN, G., LORINCZ, K., WELSH, M., MARCILLO, O., JOHNSON, J., RUIZ, M., AND LEES, J.2006. Deploying a wireless sensor network on an active volcano. IEEE Internet Computing, 18–25.


a a high frequency sampling monitoring system for

Documents